ParentFull threadtotoglazer·How, if at all, do you have to tweak model architecture or hyperparameters for pods vs tpu vs gpu?View on HN